learning/annotated_deep_learning_paper_implementations
1
0
Fork 0
mirror of https://github.com/labmlai/annotated_deep_learning_paper_implementations.git synced 2025-08-26 16:50:39 +08:00
Code Issues Packages Projects Releases Wiki Activity

LoRA notes

Browse Source
This commit is contained in:
Varuna Jayasiri
2024-08-03 16:59:15 +05:30
parent eb9337e949
commit d4af40b595
9 changed files with 1236 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
docs/RWKV/configs.html
View File

@ -12,7 +12,7 @@
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/RWKV/configs.html"/>
<meta property="og:url" content="https://nn.labml.ai/rwkv/configs.html"/>
<meta property="og:title" content="configs.py"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="configs.py"/>
@ -23,7 +23,7 @@
<title>configs.py</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/RWKV/configs.html"/>
<link rel="canonical" href="https://nn.labml.ai/rwkv/configs.html"/>
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.18/dist/katex.min.css" integrity="sha384-zTROYFVGOfTw7JV7KUu8udsvW2fx4lWOsCEDqhBreBwlHI4ioVRtmIvEThzJHGET" crossorigin="anonymous">
<!-- Global site tag (gtag.js) - Google Analytics -->
@ -47,7 +47,7 @@
<div class='docs'>
<p>
<a class="parent" href="/">home</a>
<a class="parent" href="index.html">RWKV</a>
<a class="parent" href="index.html">rwkv</a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations" target="_blank">
@ -60,7 +60,7 @@
style="max-width:100%;"/></a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/RWKV/configs.py" target="_blank">
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/rwkv/configs.py" target="_blank">
View code on Github</a>
</p>
</div>

14
docs/RWKV/experiment.html
View File

@ -12,7 +12,7 @@
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/RWKV/experiment.html"/>
<meta property="og:url" content="https://nn.labml.ai/rwkv/experiment.html"/>
<meta property="og:title" content="experiment.py"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="experiment.py"/>
@ -23,7 +23,7 @@
<title>experiment.py</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/RWKV/experiment.html"/>
<link rel="canonical" href="https://nn.labml.ai/rwkv/experiment.html"/>
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.18/dist/katex.min.css" integrity="sha384-zTROYFVGOfTw7JV7KUu8udsvW2fx4lWOsCEDqhBreBwlHI4ioVRtmIvEThzJHGET" crossorigin="anonymous">
<!-- Global site tag (gtag.js) - Google Analytics -->
@ -47,7 +47,7 @@
<div class='docs'>
<p>
<a class="parent" href="/">home</a>
<a class="parent" href="index.html">RWKV</a>
<a class="parent" href="index.html">rwkv</a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations" target="_blank">
@ -60,7 +60,7 @@
style="max-width:100%;"/></a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/RWKV/experiment.py" target="_blank">
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/rwkv/experiment.py" target="_blank">
View code on Github</a>
</p>
</div>
@ -78,10 +78,10 @@
<span class="lineno">3</span>
<span class="lineno">4</span><span class="kn">import</span> <span class="nn">torch</span>
<span class="lineno">5</span><span class="kn">import</span> <span class="nn">torch.nn</span> <span class="k">as</span> <span class="nn">nn</span>
<span class="lineno">6</span><span class="kn">from</span> <span class="nn">labml_nn.RWKV.configs</span> <span class="kn">import</span> <span class="n">RWKVConfigs</span>
<span class="lineno">6</span><span class="kn">from</span> <span class="nn">labml_nn.rwkv.configs</span> <span class="kn">import</span> <span class="n">RWKVConfigs</span>
<span class="lineno">7</span>
<span class="lineno">8</span><span class="kn">from</span> <span class="nn">labml_nn.RWKV</span> <span class="kn">import</span> <span class="n">RWKV</span>
<span class="lineno">9</span><span class="kn">from</span> <span class="nn">labml_nn.RWKV</span> <span class="kn">import</span> <span class="n">TimeMixing</span>
<span class="lineno">8</span><span class="kn">from</span> <span class="nn">labml_nn.rwkv</span> <span class="kn">import</span> <span class="n">RWKV</span>
<span class="lineno">9</span><span class="kn">from</span> <span class="nn">labml_nn.rwkv</span> <span class="kn">import</span> <span class="n">TimeMixing</span>
<span class="lineno">10</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">experiment</span>
<span class="lineno">11</span><span class="kn">from</span> <span class="nn">labml.configs</span> <span class="kn">import</span> <span class="n">option</span>
<span class="lineno">12</span><span class="kn">from</span> <span class="nn">labml_nn.experiments.nlp_autoregression</span> <span class="kn">import</span> <span class="n">NLPAutoRegressionConfigs</span></pre></div>

8
docs/RWKV/index.html
View File

@ -12,7 +12,7 @@
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/RWKV/index.html"/>
<meta property="og:url" content="https://nn.labml.ai/rwkv/index.html"/>
<meta property="og:title" content="Receptance Weighted Key Value (RWKV)"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="Receptance Weighted Key Value (RWKV)"/>
@ -23,7 +23,7 @@
<title>Receptance Weighted Key Value (RWKV)</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/RWKV/index.html"/>
<link rel="canonical" href="https://nn.labml.ai/rwkv/index.html"/>
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.18/dist/katex.min.css" integrity="sha384-zTROYFVGOfTw7JV7KUu8udsvW2fx4lWOsCEDqhBreBwlHI4ioVRtmIvEThzJHGET" crossorigin="anonymous">
<!-- Global site tag (gtag.js) - Google Analytics -->
@ -47,7 +47,7 @@
<div class='docs'>
<p>
<a class="parent" href="/">home</a>
<a class="parent" href="index.html">RWKV</a>
<a class="parent" href="index.html">rwkv</a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations" target="_blank">
@ -60,7 +60,7 @@
style="max-width:100%;"/></a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/RWKV/__init__.py" target="_blank">
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/rwkv/__init__.py" target="_blank">
View code on Github</a>
</p>
</div>

10
docs/gan/wasserstein/index.html
View File

File diff suppressed because one or more lines are too long

378
docs/lora/gpt2.html Normal file
View File

@ -0,0 +1,378 @@
<!DOCTYPE html>
<html lang="en">
<head>
<meta http-equiv="content-type" content="text/html;charset=utf-8"/>
<meta name="viewport" content="width=device-width, initial-scale=1.0"/>
<meta name="description" content=""/>
<meta name="twitter:card" content="summary"/>
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta name="twitter:title" content="gpt2.py"/>
<meta name="twitter:description" content=""/>
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/lora/gpt2.html"/>
<meta property="og:title" content="gpt2.py"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="gpt2.py"/>
<meta property="og:type" content="object"/>
<meta property="og:title" content="gpt2.py"/>
<meta property="og:description" content=""/>
<title>gpt2.py</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/lora/gpt2.html"/>
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.18/dist/katex.min.css" integrity="sha384-zTROYFVGOfTw7JV7KUu8udsvW2fx4lWOsCEDqhBreBwlHI4ioVRtmIvEThzJHGET" crossorigin="anonymous">
<!-- Global site tag (gtag.js) - Google Analytics -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-4V3HC8HBLH"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() {
dataLayer.push(arguments);
}
gtag('js', new Date());
gtag('config', 'G-4V3HC8HBLH');
</script>
</head>
<body>
<div id='container'>
<div id="background"></div>
<div class='section'>
<div class='docs'>
<p>
<a class="parent" href="/">home</a>
<a class="parent" href="index.html">lora</a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations" target="_blank">
<img alt="Github"
src="https://img.shields.io/github/stars/labmlai/annotated_deep_learning_paper_implementations?style=social"
style="max-width:100%;"/></a>
<a href="https://twitter.com/labmlai" rel="nofollow" target="_blank">
<img alt="Twitter"
src="https://img.shields.io/twitter/follow/labmlai?style=social"
style="max-width:100%;"/></a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/lora/gpt2.py" target="_blank">
View code on Github</a>
</p>
</div>
</div>
<div class='section' id='section-0'>
<div class='docs'>
<div class='section-link'>
<a href='#section-0'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">1</span><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="lineno">2</span><span class="kn">import</span> <span class="nn">torch.nn</span> <span class="k">as</span> <span class="nn">nn</span>
<span class="lineno">3</span><span class="kn">from</span> <span class="nn">transformers</span> <span class="kn">import</span> <span class="n">AutoTokenizer</span>
<span class="lineno">4</span><span class="kn">from</span> <span class="nn">labml_nn.lora</span> <span class="kn">import</span> <span class="n">Linear</span><span class="p">,</span> <span class="n">Embedding</span>
<span class="lineno">5</span>
<span class="lineno">6</span><span class="n">tokenizer</span> <span class="o">=</span> <span class="n">AutoTokenizer</span><span class="o">.</span><span class="n">from_pretrained</span><span class="p">(</span><span class="s2">&quot;gpt2&quot;</span><span class="p">)</span>
<span class="lineno">7</span>
<span class="lineno">8</span><span class="n">config</span> <span class="o">=</span> <span class="p">{</span>
<span class="lineno">9</span> <span class="s2">&quot;layer_norm_epsilon&quot;</span><span class="p">:</span> <span class="mf">1e-05</span><span class="p">,</span>
<span class="lineno">10</span> <span class="s2">&quot;n_embd&quot;</span><span class="p">:</span> <span class="mi">768</span><span class="p">,</span>
<span class="lineno">11</span> <span class="s2">&quot;n_head&quot;</span><span class="p">:</span> <span class="mi">12</span><span class="p">,</span>
<span class="lineno">12</span> <span class="s2">&quot;n_layer&quot;</span><span class="p">:</span> <span class="mi">12</span><span class="p">,</span>
<span class="lineno">13</span> <span class="s2">&quot;n_positions&quot;</span><span class="p">:</span> <span class="mi">1024</span><span class="p">,</span>
<span class="lineno">14</span> <span class="s2">&quot;vocab_size&quot;</span><span class="p">:</span> <span class="mi">50257</span><span class="p">,</span>
<span class="lineno">15</span> <span class="s2">&quot;device&quot;</span><span class="p">:</span> <span class="s2">&quot;cuda&quot;</span>
<span class="lineno">16</span><span class="p">}</span></pre></div>
</div>
</div>
<div class='section' id='section-1'>
<div class='docs'>
<div class='section-link'>
<a href='#section-1'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">19</span><span class="k">class</span> <span class="nc">FFN</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-2'>
<div class='docs'>
<div class='section-link'>
<a href='#section-2'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">20</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">dim</span><span class="p">):</span>
<span class="lineno">21</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="lineno">22</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_fc</span> <span class="o">=</span> <span class="n">Linear</span><span class="p">(</span><span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_embd&#39;</span><span class="p">],</span> <span class="n">dim</span><span class="p">,</span> <span class="n">r</span><span class="o">=</span><span class="mi">32</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="lineno">23</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_proj</span> <span class="o">=</span> <span class="n">Linear</span><span class="p">(</span><span class="n">dim</span><span class="p">,</span> <span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_embd&#39;</span><span class="p">],</span> <span class="n">r</span><span class="o">=</span><span class="mi">32</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="lineno">24</span> <span class="bp">self</span><span class="o">.</span><span class="n">act</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">functional</span><span class="o">.</span><span class="n">gelu</span></pre></div>
</div>
</div>
<div class='section' id='section-3'>
<div class='docs'>
<div class='section-link'>
<a href='#section-3'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">26</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">):</span>
<span class="lineno">27</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_fc</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="lineno">28</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">act</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="lineno">29</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_proj</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="lineno">30</span> <span class="k">return</span> <span class="n">hidden_states</span></pre></div>
</div>
</div>
<div class='section' id='section-4'>
<div class='docs'>
<div class='section-link'>
<a href='#section-4'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">33</span><span class="k">class</span> <span class="nc">MultiHeadAttention</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-5'>
<div class='docs'>
<div class='section-link'>
<a href='#section-5'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">34</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="lineno">35</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="lineno">36</span> <span class="bp">self</span><span class="o">.</span><span class="n">embed_dim</span> <span class="o">=</span> <span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_embd&#39;</span><span class="p">]</span>
<span class="lineno">37</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_heads</span> <span class="o">=</span> <span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_head&#39;</span><span class="p">]</span>
<span class="lineno">38</span> <span class="bp">self</span><span class="o">.</span><span class="n">head_dim</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">embed_dim</span> <span class="o">//</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_heads</span>
<span class="lineno">39</span> <span class="bp">self</span><span class="o">.</span><span class="n">split_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">embed_dim</span>
<span class="lineno">40</span>
<span class="lineno">41</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_att</span> <span class="o">=</span> <span class="n">Linear</span><span class="p">(</span><span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_embd&#39;</span><span class="p">],</span> <span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_embd&#39;</span><span class="p">]</span> <span class="o">*</span> <span class="mi">3</span><span class="p">,</span> <span class="n">r</span><span class="o">=</span><span class="mi">32</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="lineno">42</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_proj</span> <span class="o">=</span> <span class="n">Linear</span><span class="p">(</span><span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_embd&#39;</span><span class="p">],</span> <span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_embd&#39;</span><span class="p">],</span> <span class="n">r</span><span class="o">=</span><span class="mi">32</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-6'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-6'>#</a>
</div>
<p> Splits hidden_size dim into attn_head_size and num_heads</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">44</span> <span class="k">def</span> <span class="nf">_split_heads</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">tensor</span><span class="p">,</span> <span class="n">num_heads</span><span class="p">,</span> <span class="n">attn_head_size</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-7'>
<div class='docs'>
<div class='section-link'>
<a href='#section-7'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">48</span> <span class="n">new_shape</span> <span class="o">=</span> <span class="n">tensor</span><span class="o">.</span><span class="n">size</span><span class="p">()[:</span><span class="o">-</span><span class="mi">1</span><span class="p">]</span> <span class="o">+</span> <span class="p">(</span><span class="n">num_heads</span><span class="p">,</span> <span class="n">attn_head_size</span><span class="p">)</span>
<span class="lineno">49</span> <span class="n">tensor</span> <span class="o">=</span> <span class="n">tensor</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="n">new_shape</span><span class="p">)</span>
<span class="lineno">50</span> <span class="k">return</span> <span class="n">tensor</span><span class="o">.</span><span class="n">permute</span><span class="p">(</span><span class="mi">0</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">3</span><span class="p">)</span> <span class="c1"># (batch, head, seq_length, head_features)</span></pre></div>
</div>
</div>
<div class='section' id='section-8'>
<div class='docs'>
<div class='section-link'>
<a href='#section-8'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">52</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">):</span>
<span class="lineno">53</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">seq_length</span><span class="p">,</span> <span class="n">_</span> <span class="o">=</span> <span class="n">hidden_states</span><span class="o">.</span><span class="n">size</span><span class="p">()</span>
<span class="lineno">54</span>
<span class="lineno">55</span> <span class="n">query</span><span class="p">,</span> <span class="n">key</span><span class="p">,</span> <span class="n">value</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_att</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">split_size</span><span class="p">,</span> <span class="n">dim</span><span class="o">=</span><span class="mi">2</span><span class="p">)</span>
<span class="lineno">56</span>
<span class="lineno">57</span> <span class="n">query</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_split_heads</span><span class="p">(</span><span class="n">query</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_heads</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">head_dim</span><span class="p">)</span>
<span class="lineno">58</span> <span class="n">key</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_split_heads</span><span class="p">(</span><span class="n">key</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_heads</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">head_dim</span><span class="p">)</span>
<span class="lineno">59</span> <span class="n">value</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_split_heads</span><span class="p">(</span><span class="n">value</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_heads</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">head_dim</span><span class="p">)</span>
<span class="lineno">60</span>
<span class="lineno">61</span> <span class="n">attn_output</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">functional</span><span class="o">.</span><span class="n">scaled_dot_product_attention</span><span class="p">(</span>
<span class="lineno">62</span> <span class="n">query</span><span class="p">,</span>
<span class="lineno">63</span> <span class="n">key</span><span class="p">,</span>
<span class="lineno">64</span> <span class="n">value</span><span class="p">,</span>
<span class="lineno">65</span> <span class="n">attn_mask</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="lineno">66</span> <span class="n">dropout_p</span><span class="o">=</span><span class="mf">0.0</span><span class="p">,</span>
<span class="lineno">67</span> <span class="n">is_causal</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># for the triangular mask</span>
<span class="lineno">68</span> <span class="p">)</span>
<span class="lineno">69</span>
<span class="lineno">70</span> <span class="n">attn_output</span> <span class="o">=</span> <span class="n">attn_output</span><span class="o">.</span><span class="n">transpose</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">)</span><span class="o">.</span><span class="n">contiguous</span><span class="p">()</span>
<span class="lineno">71</span> <span class="n">attn_output</span> <span class="o">=</span> <span class="n">attn_output</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="n">batch_size</span><span class="p">,</span> <span class="n">seq_length</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">embed_dim</span><span class="p">)</span>
<span class="lineno">72</span>
<span class="lineno">73</span> <span class="n">attn_output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">c_proj</span><span class="p">(</span><span class="n">attn_output</span><span class="p">)</span>
<span class="lineno">74</span>
<span class="lineno">75</span> <span class="k">return</span> <span class="n">attn_output</span></pre></div>
</div>
</div>
<div class='section' id='section-9'>
<div class='docs'>
<div class='section-link'>
<a href='#section-9'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">78</span><span class="k">class</span> <span class="nc">Block</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-10'>
<div class='docs'>
<div class='section-link'>
<a href='#section-10'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">79</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="lineno">80</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="lineno">81</span> <span class="bp">self</span><span class="o">.</span><span class="n">pre_norm</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">LayerNorm</span><span class="p">(</span><span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_embd&#39;</span><span class="p">],</span> <span class="n">eps</span><span class="o">=</span><span class="n">config</span><span class="p">[</span><span class="s1">&#39;layer_norm_epsilon&#39;</span><span class="p">])</span>
<span class="lineno">82</span> <span class="bp">self</span><span class="o">.</span><span class="n">attn</span> <span class="o">=</span> <span class="n">MultiHeadAttention</span><span class="p">()</span>
<span class="lineno">83</span> <span class="bp">self</span><span class="o">.</span><span class="n">post_norm</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">LayerNorm</span><span class="p">(</span><span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_embd&#39;</span><span class="p">],</span> <span class="n">eps</span><span class="o">=</span><span class="n">config</span><span class="p">[</span><span class="s1">&#39;layer_norm_epsilon&#39;</span><span class="p">])</span>
<span class="lineno">84</span> <span class="bp">self</span><span class="o">.</span><span class="n">ffn</span> <span class="o">=</span> <span class="n">FFN</span><span class="p">(</span><span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_embd&#39;</span><span class="p">]</span> <span class="o">*</span> <span class="mi">4</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-11'>
<div class='docs'>
<div class='section-link'>
<a href='#section-11'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">86</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">):</span>
<span class="lineno">87</span> <span class="n">residual</span> <span class="o">=</span> <span class="n">hidden_states</span>
<span class="lineno">88</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">pre_norm</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="lineno">89</span>
<span class="lineno">90</span> <span class="n">attn_output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">attn</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="lineno">91</span>
<span class="lineno">92</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="n">attn_output</span> <span class="o">+</span> <span class="n">residual</span>
<span class="lineno">93</span> <span class="n">residual</span> <span class="o">=</span> <span class="n">hidden_states</span>
<span class="lineno">94</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">post_norm</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="lineno">95</span> <span class="n">feed_forward_output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">ffn</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="lineno">96</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="n">feed_forward_output</span> <span class="o">+</span> <span class="n">residual</span>
<span class="lineno">97</span>
<span class="lineno">98</span> <span class="k">return</span> <span class="n">hidden_states</span></pre></div>
</div>
</div>
<div class='section' id='section-12'>
<div class='docs'>
<div class='section-link'>
<a href='#section-12'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">101</span><span class="k">class</span> <span class="nc">GPTModel</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-13'>
<div class='docs'>
<div class='section-link'>
<a href='#section-13'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">102</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="lineno">103</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="lineno">104</span>
<span class="lineno">105</span> <span class="bp">self</span><span class="o">.</span><span class="n">token_embedding</span> <span class="o">=</span> <span class="n">Embedding</span><span class="p">(</span><span class="n">config</span><span class="p">[</span><span class="s1">&#39;vocab_size&#39;</span><span class="p">],</span> <span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_embd&#39;</span><span class="p">],</span> <span class="n">r</span><span class="o">=</span><span class="mi">32</span><span class="p">)</span>
<span class="lineno">106</span> <span class="bp">self</span><span class="o">.</span><span class="n">position_embedding</span> <span class="o">=</span> <span class="n">Embedding</span><span class="p">(</span><span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_positions&#39;</span><span class="p">],</span> <span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_embd&#39;</span><span class="p">],</span> <span class="n">r</span><span class="o">=</span><span class="mi">32</span><span class="p">)</span>
<span class="lineno">107</span>
<span class="lineno">108</span> <span class="bp">self</span><span class="o">.</span><span class="n">blocks</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">ModuleList</span><span class="p">([</span><span class="n">Block</span><span class="p">()</span> <span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_layer&#39;</span><span class="p">])])</span>
<span class="lineno">109</span>
<span class="lineno">110</span> <span class="bp">self</span><span class="o">.</span><span class="n">final_norm</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">LayerNorm</span><span class="p">(</span><span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_embd&#39;</span><span class="p">],</span> <span class="n">eps</span><span class="o">=</span><span class="n">config</span><span class="p">[</span><span class="s1">&#39;layer_norm_epsilon&#39;</span><span class="p">])</span>
<span class="lineno">111</span>
<span class="lineno">112</span> <span class="bp">self</span><span class="o">.</span><span class="n">lm_head</span> <span class="o">=</span> <span class="n">Linear</span><span class="p">(</span><span class="n">config</span><span class="p">[</span><span class="s1">&#39;n_embd&#39;</span><span class="p">],</span> <span class="n">config</span><span class="p">[</span><span class="s1">&#39;vocab_size&#39;</span><span class="p">],</span> <span class="n">r</span><span class="o">=</span><span class="mi">32</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-14'>
<div class='docs'>
<div class='section-link'>
<a href='#section-14'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">114</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">input_ids</span><span class="p">):</span>
<span class="lineno">115</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">input_shape</span> <span class="o">=</span> <span class="n">input_ids</span><span class="o">.</span><span class="n">size</span><span class="p">()</span>
<span class="lineno">116</span>
<span class="lineno">117</span> <span class="n">token_embeddings</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">token_embedding</span><span class="p">(</span><span class="n">input_ids</span><span class="p">)</span> <span class="c1"># B T C</span>
<span class="lineno">118</span> <span class="n">position_ids</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="n">input_shape</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="n">config</span><span class="p">[</span><span class="s1">&#39;device&#39;</span><span class="p">])</span> <span class="c1"># T C</span>
<span class="lineno">119</span> <span class="n">position_embeddings</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">position_embedding</span><span class="p">(</span><span class="n">position_ids</span><span class="p">)</span> <span class="c1"># B T C</span>
<span class="lineno">120</span>
<span class="lineno">121</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="n">token_embeddings</span> <span class="o">+</span> <span class="n">position_embeddings</span>
<span class="lineno">122</span>
<span class="lineno">123</span> <span class="k">for</span> <span class="n">block</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">blocks</span><span class="p">:</span>
<span class="lineno">124</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="n">block</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="lineno">125</span>
<span class="lineno">126</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">final_norm</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="lineno">127</span>
<span class="lineno">128</span> <span class="n">logits</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">lm_head</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="lineno">129</span>
<span class="lineno">130</span> <span class="k">return</span> <span class="n">logits</span></pre></div>
</div>
</div>
<div class='footer'>
<a href="https://labml.ai">labml.ai</a>
</div>
</div>
<script src=../interactive.js?v=1"></script>
<script>
function handleImages() {
var images = document.querySelectorAll('p>img')
for (var i = 0; i < images.length; ++i) {
handleImage(images[i])
}
}
function handleImage(img) {
img.parentElement.style.textAlign = 'center'
var modal = document.createElement('div')
modal.id = 'modal'
var modalContent = document.createElement('div')
modal.appendChild(modalContent)
var modalImage = document.createElement('img')
modalContent.appendChild(modalImage)
var span = document.createElement('span')
span.classList.add('close')
span.textContent = 'x'
modal.appendChild(span)
img.onclick = function () {
console.log('clicked')
document.body.appendChild(modal)
modalImage.src = img.src
}
span.onclick = function () {
document.body.removeChild(modal)
}
}
handleImages()
</script>
</body>
</html>

534
docs/lora/index.html Normal file
View File

@ -0,0 +1,534 @@
<!DOCTYPE html>
<html lang="en">
<head>
<meta http-equiv="content-type" content="text/html;charset=utf-8"/>
<meta name="viewport" content="width=device-width, initial-scale=1.0"/>
<meta name="description" content="Annotated implementation of RoRA from paper LoRA: Low-Rank Adaptation of Large Language Models"/>
<meta name="twitter:card" content="summary"/>
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta name="twitter:title" content="Low-Rank Adaptation (LoRA)"/>
<meta name="twitter:description" content="Annotated implementation of RoRA from paper LoRA: Low-Rank Adaptation of Large Language Models"/>
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/lora/index.html"/>
<meta property="og:title" content="Low-Rank Adaptation (LoRA)"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="Low-Rank Adaptation (LoRA)"/>
<meta property="og:type" content="object"/>
<meta property="og:title" content="Low-Rank Adaptation (LoRA)"/>
<meta property="og:description" content="Annotated implementation of RoRA from paper LoRA: Low-Rank Adaptation of Large Language Models"/>
<title>Low-Rank Adaptation (LoRA)</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/lora/index.html"/>
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.18/dist/katex.min.css" integrity="sha384-zTROYFVGOfTw7JV7KUu8udsvW2fx4lWOsCEDqhBreBwlHI4ioVRtmIvEThzJHGET" crossorigin="anonymous">
<!-- Global site tag (gtag.js) - Google Analytics -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-4V3HC8HBLH"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() {
dataLayer.push(arguments);
}
gtag('js', new Date());
gtag('config', 'G-4V3HC8HBLH');
</script>
</head>
<body>
<div id='container'>
<div id="background"></div>
<div class='section'>
<div class='docs'>
<p>
<a class="parent" href="/">home</a>
<a class="parent" href="index.html">lora</a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations" target="_blank">
<img alt="Github"
src="https://img.shields.io/github/stars/labmlai/annotated_deep_learning_paper_implementations?style=social"
style="max-width:100%;"/></a>
<a href="https://twitter.com/labmlai" rel="nofollow" target="_blank">
<img alt="Twitter"
src="https://img.shields.io/twitter/follow/labmlai?style=social"
style="max-width:100%;"/></a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/lora/__init__.py" target="_blank">
View code on Github</a>
</p>
</div>
</div>
<div class='section' id='section-0'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-0'>#</a>
</div>
<h1>Low-Rank Adaptation (LoRA)</h1>
<p>This is an implementation of <a href="https://arxiv.org/abs/2106.09685">Low-Rank Adaptation (LoRA)</a> in <a href="https://pytorch.org">PyTorch</a>.</p>
<p>Low-Rank Adaptation (LoRA) freezes pre-trained model weights and injects trainable rank decomposition matrices into each layer of the transformer. This makes it possible to efficiently fine-tune large langauge models by reducing trainable parameters by a large factor.</p>
<p>Here&#x27;s <a href="experiment.html">the training code</a> for training a GPT2 model with LoRA on Tiny Shakespeare dataset.</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">24</span><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="lineno">25</span><span class="kn">import</span> <span class="nn">torch.nn</span> <span class="k">as</span> <span class="nn">nn</span></pre></div>
</div>
</div>
<div class='section' id='section-1'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-1'>#</a>
</div>
<h2>LoRA Linear Layer</h2>
<p>LoRA linear layer adds a low-rank decomposition to the pre-trained weight matrix (<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.9991079999999999em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqc" style=""><span class="mord" style=""><span class="mord coloredeq eqq" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">d</span><span class="mbin mtight" style="">×</span><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span></span></span></span></span></span></span></span></span></span></span>) of the linear layer.</p>
<p><span ><span class="katex-display"><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.83333em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqf" style=""><span class="mord" style=""><span class="mord coloredeq eqq" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin" style="">+</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mord" style=""><span class="mord coloredeq eqo" style="">Δ</span><span class="mord mathnormal coloredeq eqo" style="margin-right:0.13889em">W</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord coloredeq eqq" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin" style="">+</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mord" style=""><span class="mord mathnormal coloredeq equ" style="margin-right:0.05017em">B</span></span><span class="mord" style=""><span class="mord mathnormal coloredeq eqt" style="">A</span></span></span></span></span></span></span></span></p>
<p>, where <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8882079999999999em;vertical-align:-0.0391em;"></span><span class="mord coloredeq eqe" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq equ" style="margin-right:0.05017em">B</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">d</span><span class="mbin mtight" style="">×</span><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span>, <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8882079999999999em;vertical-align:-0.0391em;"></span><span class="mord coloredeq eqd" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqt" style="">A</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span><span class="mbin mtight" style="">×</span><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span></span></span></span></span></span></span></span></span></span></span>, and the rank <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.5782em;vertical-align:-0.0391em;"></span><span class="mord coloredeq eqv" style=""><span class="mord mathnormal" style="margin-right:0.02778em">r</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel">≪</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span></span><span class="base"><span class="strut" style="height:1em;vertical-align:-0.25em;"></span><span class="mord mathnormal">min</span><span class="mopen">(</span><span class="mord mathnormal">d</span><span class="mpunct">,</span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mord mathnormal" style="margin-right:0.03148em;">k</span><span class="mclose">)</span></span></span></span></span>.</p>
<p>All parameters are frozen except <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqt" style=""><span class="mord mathnormal" style="">A</span></span></span></span></span></span> and <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq equ" style=""><span class="mord mathnormal" style="margin-right:0.05017em">B</span></span></span></span></span></span>.</p>
<p><span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqo" style=""><span class="mord" style="">Δ</span><span class="mord mathnormal" style="margin-right:0.13889em">W</span></span></span></span></span></span> is initialized to be zero at the beginning of the training.</p>
<p>They multiple <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqm" style=""><span class="mord" style=""><span class="mord coloredeq eqo" style="">Δ</span><span class="mord mathnormal coloredeq eqo" style="margin-right:0.13889em">W</span></span><span class="mord mathnormal" style="">x</span></span></span></span></span></span> by <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.040392em;vertical-align:-0.345em;"></span><span class="mord coloredeq eqi" style=""><span class="mord" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.695392em;"><span style="top:-2.6550000000000002em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.394em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqp" style="margin-right:0.0037em">α</span></span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.345em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span></span></span></span></span> where <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqp" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span></span></span></span></span> is a hyper-parameter. Once <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqp" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span></span></span></span></span> is tuned it can be kept the same when varying <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqv" style=""><span class="mord mathnormal" style="margin-right:0.02778em">r</span></span></span></span></span></span>.</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">28</span><span class="k">class</span> <span class="nc">Linear</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-2'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-2'>#</a>
</div>
<ul><li><code class="highlight"><span></span><span class="n">in_features</span></code>
is the number of input features of the linear layer </li>
<li><code class="highlight"><span></span><span class="n">out_features</span></code>
is the number of output features of the linear layer </li>
<li><code class="highlight"><span></span><span class="n">bias</span></code>
is a flag indicating if there is a bias parameter </li>
<li><code class="highlight"><span></span><span class="n">r</span></code>
is the rank of the decomposition <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqv" style=""><span class="mord mathnormal" style="margin-right:0.02778em">r</span></span></span></span></span></span> </li>
<li><code class="highlight"><span></span><span class="n">alpha</span></code>
is the scaling factor <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqp" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span></span></span></span></span></li></ul>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">49</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">in_features</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">out_features</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">bias</span><span class="p">:</span> <span class="nb">bool</span><span class="p">,</span>
<span class="lineno">50</span> <span class="n">r</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">alpha</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-3'>
<div class='docs'>
<div class='section-link'>
<a href='#section-3'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">58</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span></pre></div>
</div>
</div>
<div class='section' id='section-4'>
<div class='docs'>
<div class='section-link'>
<a href='#section-4'>#</a>
</div>
<p>Set <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqn" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqp" style="margin-right:0.0037em">α</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathnormal coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span></span></span> is not provided. i.e. make the scaling factor <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.040392em;vertical-align:-0.345em;"></span><span class="mord coloredeq eqh" style=""><span class="mord" style=""><span class="mord coloredeq eqi" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.695392em;"><span style="top:-2.6550000000000002em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.394em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqp" style="margin-right:0.0037em">α</span></span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.345em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style="">1</span></span></span></span></span></span>. </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">61</span> <span class="k">if</span> <span class="n">alpha</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="lineno">62</span> <span class="n">alpha</span> <span class="o">=</span> <span class="n">r</span></pre></div>
</div>
</div>
<div class='section' id='section-5'>
<div class='docs'>
<div class='section-link'>
<a href='#section-5'>#</a>
</div>
<p>The pre-trained weight <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.83333em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqq" style=""><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span></span> </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">65</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">empty</span><span class="p">((</span><span class="n">out_features</span><span class="p">,</span> <span class="n">in_features</span><span class="p">)))</span></pre></div>
</div>
</div>
<div class='section' id='section-6'>
<div class='docs'>
<div class='section-link'>
<a href='#section-6'>#</a>
</div>
<p>Freeze it </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">67</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight</span><span class="o">.</span><span class="n">requires_grad</span> <span class="o">=</span> <span class="kc">False</span>
<span class="lineno">68</span>
<span class="lineno">69</span> <span class="k">if</span> <span class="n">bias</span><span class="p">:</span></pre></div>
</div>
</div>
<div class='section' id='section-7'>
<div class='docs'>
<div class='section-link'>
<a href='#section-7'>#</a>
</div>
<p>Bias parameter <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.84444em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqr" style=""><span class="mord" style=""><span class="mord mathnormal" style="">b</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:0em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span></span> (also frozen) </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">71</span> <span class="bp">self</span><span class="o">.</span><span class="n">bias</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">empty</span><span class="p">(</span><span class="n">out_features</span><span class="p">))</span>
<span class="lineno">72</span> <span class="bp">self</span><span class="o">.</span><span class="n">bias</span><span class="o">.</span><span class="n">requires_grad</span> <span class="o">=</span> <span class="kc">False</span>
<span class="lineno">73</span> <span class="k">else</span><span class="p">:</span></pre></div>
</div>
</div>
<div class='section' id='section-8'>
<div class='docs'>
<div class='section-link'>
<a href='#section-8'>#</a>
</div>
<p>No bias parameter </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">75</span> <span class="bp">self</span><span class="o">.</span><span class="n">bias</span> <span class="o">=</span> <span class="kc">None</span></pre></div>
</div>
</div>
<div class='section' id='section-9'>
<div class='docs'>
<div class='section-link'>
<a href='#section-9'>#</a>
</div>
<p>scaling factor <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.040392em;vertical-align:-0.345em;"></span><span class="mord coloredeq eqi" style=""><span class="mord" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.695392em;"><span style="top:-2.6550000000000002em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.394em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqp" style="margin-right:0.0037em">α</span></span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.345em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span></span></span></span></span> </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">78</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaling</span> <span class="o">=</span> <span class="n">alpha</span> <span class="o">/</span> <span class="n">r</span></pre></div>
</div>
</div>
<div class='section' id='section-10'>
<div class='docs'>
<div class='section-link'>
<a href='#section-10'>#</a>
</div>
<p>Matrix <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8882079999999999em;vertical-align:-0.0391em;"></span><span class="mord coloredeq eqd" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqt" style="">A</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span><span class="mbin mtight" style="">×</span><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span></span></span></span></span></span></span></span></span></span></span> </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">80</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_a</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">empty</span><span class="p">((</span><span class="n">in_features</span><span class="p">,</span> <span class="n">r</span><span class="p">)))</span></pre></div>
</div>
</div>
<div class='section' id='section-11'>
<div class='docs'>
<div class='section-link'>
<a href='#section-11'>#</a>
</div>
<p>Matrix <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8882079999999999em;vertical-align:-0.0391em;"></span><span class="mord coloredeq eqe" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq equ" style="margin-right:0.05017em">B</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">d</span><span class="mbin mtight" style="">×</span><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span>, we keep <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqt" style=""><span class="mord mathnormal" style="">A</span></span></span></span></span></span> and <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq equ" style=""><span class="mord mathnormal" style="margin-right:0.05017em">B</span></span></span></span></span></span> transposed </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">82</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_b</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">empty</span><span class="p">((</span><span class="n">r</span><span class="p">,</span> <span class="n">out_features</span><span class="p">)))</span>
<span class="lineno">83</span>
<span class="lineno">84</span> <span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span></pre></div>
</div>
</div>
<div class='section' id='section-12'>
<div class='docs'>
<div class='section-link'>
<a href='#section-12'>#</a>
</div>
<p>Initialize <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqt" style=""><span class="mord mathnormal" style="">A</span></span></span></span></span></span> similar to a weight matrix in a normal linear layer </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">86</span> <span class="n">nn</span><span class="o">.</span><span class="n">init</span><span class="o">.</span><span class="n">kaiming_uniform_</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">lora_a</span><span class="p">,</span> <span class="n">a</span><span class="o">=</span><span class="mi">5</span> <span class="o">**</span> <span class="mf">0.5</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-13'>
<div class='docs'>
<div class='section-link'>
<a href='#section-13'>#</a>
</div>
<p>Initialize <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq equ" style=""><span class="mord mathnormal" style="margin-right:0.05017em">B</span></span></span></span></span></span> to <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.64444em;vertical-align:0em;"></span><span class="mord coloredeq eqs" style=""><span class="mord" style="">0</span></span></span></span></span></span> so that <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqk" style=""><span class="mord" style=""><span class="mord coloredeq eqo" style="">Δ</span><span class="mord mathnormal coloredeq eqo" style="margin-right:0.13889em">W</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathnormal coloredeq equ" style="margin-right:0.05017em">B</span></span><span class="mord" style=""><span class="mord mathnormal coloredeq eqt" style="">A</span></span></span></span></span></span></span> is <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.64444em;vertical-align:0em;"></span><span class="mord coloredeq eqs" style=""><span class="mord" style="">0</span></span></span></span></span></span> at initialization </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">88</span> <span class="n">nn</span><span class="o">.</span><span class="n">init</span><span class="o">.</span><span class="n">zeros_</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">lora_b</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-14'>
<div class='docs'>
<div class='section-link'>
<a href='#section-14'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">90</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-15'>
<div class='docs'>
<div class='section-link'>
<a href='#section-15'>#</a>
</div>
<p>Compute <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.83333em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqq" style=""><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mord mathnormal">x</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin">+</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span></span><span class="base"><span class="strut" style="height:0.84444em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqr" style=""><span class="mord" style=""><span class="mord mathnormal" style="">b</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:0em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span></span> </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">92</span> <span class="n">result</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">functional</span><span class="o">.</span><span class="n">linear</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">bias</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-16'>
<div class='docs'>
<div class='section-link'>
<a href='#section-16'>#</a>
</div>
<p>Add <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.040392em;vertical-align:-0.345em;"></span><span class="mord coloredeq eqi" style=""><span class="mord" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.695392em;"><span style="top:-2.6550000000000002em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.394em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqp" style="margin-right:0.0037em">α</span></span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.345em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span><span class="mord coloredeq eqm" style=""><span class="mord" style=""><span class="mord coloredeq eqo" style="">Δ</span><span class="mord mathnormal coloredeq eqo" style="margin-right:0.13889em">W</span></span><span class="mord mathnormal" style="">x</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span></span><span class="base"><span class="strut" style="height:1.040392em;vertical-align:-0.345em;"></span><span class="mord coloredeq eqi" style=""><span class="mord" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.695392em;"><span style="top:-2.6550000000000002em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.394em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqp" style="margin-right:0.0037em">α</span></span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.345em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span><span class="mord coloredeq equ" style=""><span class="mord mathnormal" style="margin-right:0.05017em">B</span></span><span class="mord coloredeq eqt" style=""><span class="mord mathnormal" style="">A</span></span><span class="mord mathnormal">x</span></span></span></span></span> </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">95</span> <span class="n">result</span> <span class="o">+=</span> <span class="p">(</span><span class="n">x</span> <span class="o">@</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_a</span> <span class="o">@</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_b</span><span class="p">)</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaling</span></pre></div>
</div>
</div>
<div class='section' id='section-17'>
<div class='docs'>
<div class='section-link'>
<a href='#section-17'>#</a>
</div>
<p> </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">98</span> <span class="k">return</span> <span class="n">result</span></pre></div>
</div>
</div>
<div class='section' id='section-18'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-18'>#</a>
</div>
<h2>LoRA Embedding Layer</h2>
<p>Similar to LoRA linear layer this adds a low-rank decomposition to the pre-trained embedding weights matrix (<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.9991079999999999em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqc" style=""><span class="mord" style=""><span class="mord coloredeq eqq" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">d</span><span class="mbin mtight" style="">×</span><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span></span></span></span></span></span></span></span></span></span></span>).</p>
<p><span ><span class="katex-display"><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.83333em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqf" style=""><span class="mord" style=""><span class="mord coloredeq eqq" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin" style="">+</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mord" style=""><span class="mord coloredeq eqo" style="">Δ</span><span class="mord mathnormal coloredeq eqo" style="margin-right:0.13889em">W</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord coloredeq eqq" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin" style="">+</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mord" style=""><span class="mord mathnormal coloredeq equ" style="margin-right:0.05017em">B</span></span><span class="mord" style=""><span class="mord mathnormal coloredeq eqt" style="">A</span></span></span></span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">101</span><span class="k">class</span> <span class="nc">Embedding</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-19'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-19'>#</a>
</div>
<ul><li><code class="highlight"><span></span><span class="n">num_embeddings</span></code>
is the number of embeddings </li>
<li><code class="highlight"><span></span><span class="n">embedding_dim</span></code>
is the number embedding dimensions </li>
<li><code class="highlight"><span></span><span class="n">r</span></code>
is the rank of the decomposition <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqv" style=""><span class="mord mathnormal" style="margin-right:0.02778em">r</span></span></span></span></span></span> </li>
<li><code class="highlight"><span></span><span class="n">alpha</span></code>
is the scaling factor <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqp" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span></span></span></span></span></li></ul>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">111</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">num_embeddings</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">embedding_dim</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span>
<span class="lineno">112</span> <span class="n">r</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">alpha</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-20'>
<div class='docs'>
<div class='section-link'>
<a href='#section-20'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">120</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span></pre></div>
</div>
</div>
<div class='section' id='section-21'>
<div class='docs'>
<div class='section-link'>
<a href='#section-21'>#</a>
</div>
<p>Set <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqn" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqp" style="margin-right:0.0037em">α</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathnormal coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span></span></span> is not provided. i.e. make the scaling factor <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.040392em;vertical-align:-0.345em;"></span><span class="mord coloredeq eqh" style=""><span class="mord" style=""><span class="mord coloredeq eqi" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.695392em;"><span style="top:-2.6550000000000002em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.394em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqp" style="margin-right:0.0037em">α</span></span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.345em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style="">1</span></span></span></span></span></span>. </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">123</span> <span class="k">if</span> <span class="n">alpha</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="lineno">124</span> <span class="n">alpha</span> <span class="o">=</span> <span class="n">r</span></pre></div>
</div>
</div>
<div class='section' id='section-22'>
<div class='docs'>
<div class='section-link'>
<a href='#section-22'>#</a>
</div>
<p>The pre-trained embedding weights <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.83333em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqq" style=""><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span></span> (frozen) </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">127</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">empty</span><span class="p">((</span><span class="n">num_embeddings</span><span class="p">,</span> <span class="n">embedding_dim</span><span class="p">)))</span>
<span class="lineno">128</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight</span><span class="o">.</span><span class="n">requires_grad</span> <span class="o">=</span> <span class="kc">False</span></pre></div>
</div>
</div>
<div class='section' id='section-23'>
<div class='docs'>
<div class='section-link'>
<a href='#section-23'>#</a>
</div>
<p>scaling factor <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.040392em;vertical-align:-0.345em;"></span><span class="mord coloredeq eqi" style=""><span class="mord" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.695392em;"><span style="top:-2.6550000000000002em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.394em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqp" style="margin-right:0.0037em">α</span></span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.345em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span></span></span></span></span> </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">131</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaling</span> <span class="o">=</span> <span class="n">alpha</span> <span class="o">/</span> <span class="n">r</span></pre></div>
</div>
</div>
<div class='section' id='section-24'>
<div class='docs'>
<div class='section-link'>
<a href='#section-24'>#</a>
</div>
<p>Matrix <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8882079999999999em;vertical-align:-0.0391em;"></span><span class="mord coloredeq eqd" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqt" style="">A</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span><span class="mbin mtight" style="">×</span><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span></span></span></span></span></span></span></span></span></span></span> </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">133</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_a</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">empty</span><span class="p">((</span><span class="n">num_embeddings</span><span class="p">,</span> <span class="n">r</span><span class="p">)))</span></pre></div>
</div>
</div>
<div class='section' id='section-25'>
<div class='docs'>
<div class='section-link'>
<a href='#section-25'>#</a>
</div>
<p>Matrix <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8882079999999999em;vertical-align:-0.0391em;"></span><span class="mord coloredeq eqe" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq equ" style="margin-right:0.05017em">B</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">∈</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathbb" style="">R</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8491079999999999em;"><span style="top:-3.063em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">d</span><span class="mbin mtight" style="">×</span><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqv" style="margin-right:0.02778em">r</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span> </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">135</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_b</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">empty</span><span class="p">((</span><span class="n">r</span><span class="p">,</span> <span class="n">embedding_dim</span><span class="p">)))</span>
<span class="lineno">136</span>
<span class="lineno">137</span> <span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span></pre></div>
</div>
</div>
<div class='section' id='section-26'>
<div class='docs'>
<div class='section-link'>
<a href='#section-26'>#</a>
</div>
<p>Initialize <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqt" style=""><span class="mord mathnormal" style="">A</span></span></span></span></span></span> with a normal distribution </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">139</span> <span class="n">nn</span><span class="o">.</span><span class="n">init</span><span class="o">.</span><span class="n">normal_</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">lora_a</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-27'>
<div class='docs'>
<div class='section-link'>
<a href='#section-27'>#</a>
</div>
<p>Initialize <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq equ" style=""><span class="mord mathnormal" style="margin-right:0.05017em">B</span></span></span></span></span></span> to <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.64444em;vertical-align:0em;"></span><span class="mord coloredeq eqs" style=""><span class="mord" style="">0</span></span></span></span></span></span> so that <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord coloredeq eqk" style=""><span class="mord" style=""><span class="mord coloredeq eqo" style="">Δ</span><span class="mord mathnormal coloredeq eqo" style="margin-right:0.13889em">W</span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel" style="">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mord" style=""><span class="mord mathnormal coloredeq equ" style="margin-right:0.05017em">B</span></span><span class="mord" style=""><span class="mord mathnormal coloredeq eqt" style="">A</span></span></span></span></span></span></span> is <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.64444em;vertical-align:0em;"></span><span class="mord coloredeq eqs" style=""><span class="mord" style="">0</span></span></span></span></span></span> at initialization </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">141</span> <span class="n">nn</span><span class="o">.</span><span class="n">init</span><span class="o">.</span><span class="n">zeros_</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">lora_b</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-28'>
<div class='docs'>
<div class='section-link'>
<a href='#section-28'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">143</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-29'>
<div class='docs'>
<div class='section-link'>
<a href='#section-29'>#</a>
</div>
<p>Compute the embeddings <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1em;vertical-align:-0.25em;"></span><span class="mord coloredeq eqq" style=""><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.13889em">W</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.13889em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight coloredeq eqs" style="">0</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span><span class="mord text"><span class="mord">onehot</span></span><span class="mopen">(</span><span class="mord mathnormal">x</span><span class="mclose">)</span></span></span></span></span> </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">145</span> <span class="n">result</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">functional</span><span class="o">.</span><span class="n">embedding</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-30'>
<div class='docs'>
<div class='section-link'>
<a href='#section-30'>#</a>
</div>
<p>Add <span ><strong style="">Error</strong></span> </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">148</span> <span class="n">result</span> <span class="o">+=</span> <span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">functional</span><span class="o">.</span><span class="n">embedding</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_a</span><span class="p">)</span> <span class="o">@</span> <span class="bp">self</span><span class="o">.</span><span class="n">lora_b</span><span class="p">)</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaling</span></pre></div>
</div>
</div>
<div class='section' id='section-31'>
<div class='docs'>
<div class='section-link'>
<a href='#section-31'>#</a>
</div>
<p> </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">151</span> <span class="k">return</span> <span class="n">result</span></pre></div>
</div>
</div>
<div class='footer'>
<a href="https://labml.ai">labml.ai</a>
</div>
</div>
<script src=../interactive.js?v=1"></script>
<script>
function handleImages() {
var images = document.querySelectorAll('p>img')
for (var i = 0; i < images.length; ++i) {
handleImage(images[i])
}
}
function handleImage(img) {
img.parentElement.style.textAlign = 'center'
var modal = document.createElement('div')
modal.id = 'modal'
var modalContent = document.createElement('div')
modal.appendChild(modalContent)
var modalImage = document.createElement('img')
modalContent.appendChild(modalImage)
var span = document.createElement('span')
span.classList.add('close')
span.textContent = 'x'
modal.appendChild(span)
img.onclick = function () {
console.log('clicked')
document.body.appendChild(modal)
modalImage.src = img.src
}
span.onclick = function () {
document.body.removeChild(modal)
}
}
handleImages()
</script>
</body>
</html>

186
docs/lora/transform_hf_model.html Normal file
View File

@ -0,0 +1,186 @@
<!DOCTYPE html>
<html lang="en">
<head>
<meta http-equiv="content-type" content="text/html;charset=utf-8"/>
<meta name="viewport" content="width=device-width, initial-scale=1.0"/>
<meta name="description" content=""/>
<meta name="twitter:card" content="summary"/>
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta name="twitter:title" content="transform_hf_model.py"/>
<meta name="twitter:description" content=""/>
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/lora/transform_hf_model.html"/>
<meta property="og:title" content="transform_hf_model.py"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="transform_hf_model.py"/>
<meta property="og:type" content="object"/>
<meta property="og:title" content="transform_hf_model.py"/>
<meta property="og:description" content=""/>
<title>transform_hf_model.py</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/lora/transform_hf_model.html"/>
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.18/dist/katex.min.css" integrity="sha384-zTROYFVGOfTw7JV7KUu8udsvW2fx4lWOsCEDqhBreBwlHI4ioVRtmIvEThzJHGET" crossorigin="anonymous">
<!-- Global site tag (gtag.js) - Google Analytics -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-4V3HC8HBLH"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() {
dataLayer.push(arguments);
}
gtag('js', new Date());
gtag('config', 'G-4V3HC8HBLH');
</script>
</head>
<body>
<div id='container'>
<div id="background"></div>
<div class='section'>
<div class='docs'>
<p>
<a class="parent" href="/">home</a>
<a class="parent" href="index.html">lora</a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations" target="_blank">
<img alt="Github"
src="https://img.shields.io/github/stars/labmlai/annotated_deep_learning_paper_implementations?style=social"
style="max-width:100%;"/></a>
<a href="https://twitter.com/labmlai" rel="nofollow" target="_blank">
<img alt="Twitter"
src="https://img.shields.io/twitter/follow/labmlai?style=social"
style="max-width:100%;"/></a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/lora/transform_hf_model.py" target="_blank">
View code on Github</a>
</p>
</div>
</div>
<div class='section' id='section-0'>
<div class='docs'>
<div class='section-link'>
<a href='#section-0'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">1</span><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="lineno">2</span><span class="kn">from</span> <span class="nn">transformers</span> <span class="kn">import</span> <span class="n">AutoModelForCausalLM</span></pre></div>
</div>
</div>
<div class='section' id='section-1'>
<div class='docs'>
<div class='section-link'>
<a href='#section-1'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">5</span><span class="k">def</span> <span class="nf">transform_hf_model</span><span class="p">():</span>
<span class="lineno">6</span> <span class="n">model</span> <span class="o">=</span> <span class="n">AutoModelForCausalLM</span><span class="o">.</span><span class="n">from_pretrained</span><span class="p">(</span><span class="s2">&quot;gpt2&quot;</span><span class="p">)</span>
<span class="lineno">7</span>
<span class="lineno">8</span> <span class="n">state_dict</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">state_dict</span><span class="p">()</span>
<span class="lineno">9</span>
<span class="lineno">10</span> <span class="n">mapping</span> <span class="o">=</span> <span class="p">{</span>
<span class="lineno">11</span> <span class="s1">&#39;transformer.wte.weight&#39;</span><span class="p">:</span> <span class="s1">&#39;token_embedding.weight&#39;</span><span class="p">,</span>
<span class="lineno">12</span> <span class="s1">&#39;transformer.wpe.weight&#39;</span><span class="p">:</span> <span class="s1">&#39;position_embedding.weight&#39;</span><span class="p">,</span>
<span class="lineno">13</span> <span class="s1">&#39;transformer.ln_f.weight&#39;</span><span class="p">:</span> <span class="s1">&#39;final_norm.weight&#39;</span><span class="p">,</span>
<span class="lineno">14</span> <span class="s1">&#39;transformer.ln_f.bias&#39;</span><span class="p">:</span> <span class="s1">&#39;final_norm.bias&#39;</span><span class="p">,</span>
<span class="lineno">15</span> <span class="s1">&#39;lm_head.weight&#39;</span><span class="p">:</span> <span class="s1">&#39;lm_head.weight&#39;</span>
<span class="lineno">16</span> <span class="p">}</span>
<span class="lineno">17</span>
<span class="lineno">18</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">12</span><span class="p">):</span>
<span class="lineno">19</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">&#39;transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ln_1.weight&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.pre_norm.weight&#39;</span>
<span class="lineno">20</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">&#39;transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ln_1.bias&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.pre_norm.bias&#39;</span>
<span class="lineno">21</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">&#39;transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_attn.weight&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_att.weight&#39;</span>
<span class="lineno">22</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">&#39;transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_attn.bias&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_att.bias&#39;</span>
<span class="lineno">23</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">&#39;transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_proj.weight&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_proj.weight&#39;</span>
<span class="lineno">24</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">&#39;transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_proj.bias&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_proj.bias&#39;</span>
<span class="lineno">25</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">&#39;transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ln_2.weight&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.post_norm.weight&#39;</span>
<span class="lineno">26</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">&#39;transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ln_2.bias&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.post_norm.bias&#39;</span>
<span class="lineno">27</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">&#39;transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.mlp.c_fc.weight&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ffn.c_fc.weight&#39;</span>
<span class="lineno">28</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">&#39;transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.mlp.c_fc.bias&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ffn.c_fc.bias&#39;</span>
<span class="lineno">29</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">&#39;transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.mlp.c_proj.weight&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ffn.c_proj.weight&#39;</span>
<span class="lineno">30</span> <span class="n">mapping</span><span class="p">[</span><span class="sa">f</span><span class="s1">&#39;transformer.h.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.mlp.c_proj.bias&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ffn.c_proj.bias&#39;</span>
<span class="lineno">31</span>
<span class="lineno">32</span> <span class="n">new_state_dict</span> <span class="o">=</span> <span class="p">{}</span>
<span class="lineno">33</span> <span class="k">for</span> <span class="n">old_key</span><span class="p">,</span> <span class="n">new_key</span> <span class="ow">in</span> <span class="n">mapping</span><span class="o">.</span><span class="n">items</span><span class="p">():</span>
<span class="lineno">34</span> <span class="k">if</span> <span class="n">old_key</span> <span class="ow">in</span> <span class="n">state_dict</span><span class="p">:</span>
<span class="lineno">35</span> <span class="n">new_state_dict</span><span class="p">[</span><span class="n">new_key</span><span class="p">]</span> <span class="o">=</span> <span class="n">state_dict</span><span class="p">[</span><span class="n">old_key</span><span class="p">]</span></pre></div>
</div>
</div>
<div class='section' id='section-2'>
<div class='docs'>
<div class='section-link'>
<a href='#section-2'>#</a>
</div>
<p>transpose weight matrices of convo 1d layers to use linear layers instead </p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">38</span> <span class="n">convo_layers</span> <span class="o">=</span> <span class="p">([</span><span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ffn.c_fc.weight&#39;</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">12</span><span class="p">)]</span> <span class="o">+</span>
<span class="lineno">39</span> <span class="p">[</span><span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.ffn.c_proj.weight&#39;</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">12</span><span class="p">)]</span> <span class="o">+</span>
<span class="lineno">40</span> <span class="p">[</span><span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_att.weight&#39;</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">12</span><span class="p">)]</span> <span class="o">+</span>
<span class="lineno">41</span> <span class="p">[</span><span class="sa">f</span><span class="s1">&#39;blocks.</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">.attn.c_proj.weight&#39;</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">12</span><span class="p">)])</span>
<span class="lineno">42</span>
<span class="lineno">43</span> <span class="k">for</span> <span class="n">layer</span> <span class="ow">in</span> <span class="n">convo_layers</span><span class="p">:</span>
<span class="lineno">44</span> <span class="n">new_state_dict</span><span class="p">[</span><span class="n">layer</span><span class="p">]</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">transpose</span><span class="p">(</span><span class="n">new_state_dict</span><span class="p">[</span><span class="n">layer</span><span class="p">],</span> <span class="mi">0</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
<span class="lineno">45</span>
<span class="lineno">46</span> <span class="n">torch</span><span class="o">.</span><span class="n">save</span><span class="p">(</span><span class="n">new_state_dict</span><span class="p">,</span> <span class="s1">&#39;transformed.pth&#39;</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='footer'>
<a href="https://labml.ai">labml.ai</a>
</div>
</div>
<script src=../interactive.js?v=1"></script>
<script>
function handleImages() {
var images = document.querySelectorAll('p>img')
for (var i = 0; i < images.length; ++i) {
handleImage(images[i])
}
}
function handleImage(img) {
img.parentElement.style.textAlign = 'center'
var modal = document.createElement('div')
modal.id = 'modal'
var modalContent = document.createElement('div')
modal.appendChild(modalContent)
var modalImage = document.createElement('img')
modalContent.appendChild(modalImage)
var span = document.createElement('span')
span.classList.add('close')
span.textContent = 'x'
modal.appendChild(span)
img.onclick = function () {
console.log('clicked')
document.body.appendChild(modal)
modalImage.src = img.src
}
span.onclick = function () {
document.body.removeChild(modal)
}
}
handleImages()
</script>
</body>
</html>

35
docs/sitemap.xml
View File

@ -8,7 +8,7 @@
<url>
<loc>https://nn.labml.ai/gan/wasserstein/index.html</loc>
<lastmod>2023-10-24T16:30:00+00:00</lastmod>
<lastmod>2024-07-15T16:30:00+00:00</lastmod>
<priority>1.00</priority>
</url>
@ -504,22 +504,22 @@
<url>
<loc>https://nn.labml.ai/RWKV/configs.html</loc>
<lastmod>2024-03-17T16:30:00+00:00</lastmod>
<loc>https://nn.labml.ai/rwkv/configs.html</loc>
<lastmod>2024-08-02T16:30:00+00:00</lastmod>
<priority>1.00</priority>
</url>
<url>
<loc>https://nn.labml.ai/RWKV/index.html</loc>
<lastmod>2024-03-17T16:30:00+00:00</lastmod>
<loc>https://nn.labml.ai/rwkv/index.html</loc>
<lastmod>2024-08-02T16:30:00+00:00</lastmod>
<priority>1.00</priority>
</url>
<url>
<loc>https://nn.labml.ai/RWKV/experiment.html</loc>
<lastmod>2024-03-17T16:30:00+00:00</lastmod>
<loc>https://nn.labml.ai/rwkv/experiment.html</loc>
<lastmod>2024-08-02T16:30:00+00:00</lastmod>
<priority>1.00</priority>
</url>
@ -1294,6 +1294,27 @@
</url>
<url>
<loc>https://nn.labml.ai/lora/gpt2.html</loc>
<lastmod>2024-08-02T16:30:00+00:00</lastmod>
<priority>1.00</priority>
</url>
<url>
<loc>https://nn.labml.ai/lora/index.html</loc>
<lastmod>2024-08-02T16:30:00+00:00</lastmod>
<priority>1.00</priority>
</url>
<url>
<loc>https://nn.labml.ai/lora/transform_hf_model.html</loc>
<lastmod>2024-08-02T16:30:00+00:00</lastmod>
<priority>1.00</priority>
</url>
<url>
<loc>https://nn.labml.ai/graphs/gat/index.html</loc>
<lastmod>2023-10-24T16:30:00+00:00</lastmod>

91
labml_nn/lora/__init__.py
View File

@ -1,5 +1,24 @@
"""
# LoRA
---
title: Low-Rank Adaptation (LoRA)
summary: >
Annotated implementation of RoRA from paper
LoRA: Low-Rank Adaptation of Large Language Models
---
# Low-Rank Adaptation (LoRA)
This is an implementation of
[Low-Rank Adaptation (LoRA)](https://arxiv.org/abs/2106.09685)
in [PyTorch](https://pytorch.org).
Low-Rank Adaptation (LoRA) freezes pre-trained model weights and injects
trainable rank decomposition matrices into each layer of the transformer.
This makes it possible to efficiently fine-tune large langauge models by
reducing trainable parameters by a large factor.
Here's [the training code](experiment.html) for training a GPT2 model with LoRA
on Tiny Shakespeare dataset.
"""
import torch
@ -7,56 +26,126 @@ import torch.nn as nn
class Linear(nn.Module):
"""
## LoRA Linear Layer
LoRA linear layer adds a low-rank decomposition to the pre-trained
weight matrix ($W_0 \in \mathbb{R}^{d \times k}$)
of the linear layer.
$$W_0 + \Delta W = W_0 + BA$$
, where $B \in \mathbb{R}^{d \times r}$, $A \in \mathbb{R}^{r \times k}$,
and the rank $r \ll min(d, k)$.
All parameters are frozen except $A$ and $B$.
$\Delta W$ is initialized to be zero at the beginning of the training.
They multiple $\Delta W x$ by $\frac{\alpha}{r}$ where $\alpha$ is a hyper-parameter.
Once $\alpha$ is tuned it can be kept the same when varying $r$.
"""
def __init__(self, in_features: int, out_features: int, bias: bool,
r: int, alpha: int = None):
"""
:param in_features: is the number of input features of the linear layer
:param out_features: is the number of output features of the linear layer
:param bias: is a flag indicating if there is a bias parameter
:param r: is the rank of the decomposition $r$
:param alpha: is the scaling factor $\alpha$
"""
super().__init__()
# Set $\alpha = r$ is not provided. i.e. make the scaling factor $\frac{\alpha}{r} = 1$.
if alpha is None:
alpha = r
# The pre-trained weight $W_0$
self.weight = nn.Parameter(torch.empty((out_features, in_features)))
# Freeze it
self.weight.requires_grad = False
if bias:
# Bias parameter $b_0$ (also frozen)
self.bias = nn.Parameter(torch.empty(out_features))
self.bias.requires_grad = False
else:
# No bias parameter
self.bias = None
# scaling factor $\frac{\alpha}{r}$
self.scaling = alpha / r
# Matrix $A \in \mathbb{R}^{r \times k}$
self.lora_a = nn.Parameter(torch.empty((in_features, r)))
# Matrix $B \in \mathbb{R}^{d \times r}$, we keep $A$ and $B$ transposed
self.lora_b = nn.Parameter(torch.empty((r, out_features)))
with torch.no_grad():
# Initialize $A$ similar to a weight matrix in a normal linear layer
nn.init.kaiming_uniform_(self.lora_a, a=5 ** 0.5)
# Initialize $B$ to $0$ so that $\Delta W = BA$ is $0$ at initialization
nn.init.zeros_(self.lora_b)
def forward(self, x: torch.Tensor):
# Compute $W_0 x + b_0$
result = nn.functional.linear(x, self.weight, bias=self.bias)
# Add $\frac{\alpha}{r} \Delta W x = \frac{\alpha}{r} BAx$
result += (x @ self.lora_a @ self.lora_b) * self.scaling
#
return result
class Embedding(nn.Module):
"""
## LoRA Embedding Layer
Similar to LoRA linear layer this adds a low-rank decomposition to the pre-trained
embedding weights matrix ($W_0 \in \mathbb{R}^{d \times k}$).
$$W_0 + \Delta W = W_0 + BA$$
"""
def __init__(self, num_embeddings: int, embedding_dim: int,
r: int, alpha: int = None):
"""
:param num_embeddings: is the number of embeddings
:param embedding_dim: is the number embedding dimensions
:param r: is the rank of the decomposition $r$
:param alpha: is the scaling factor $\alpha$
"""
super().__init__()
# Set $\alpha = r$ is not provided. i.e. make the scaling factor $\frac{\alpha}{r} = 1$.
if alpha is None:
alpha = r
# The pre-trained embedding weights $W_0$ (frozen)
self.weight = nn.Parameter(torch.empty((num_embeddings, embedding_dim)))
self.weight.requires_grad = False
# scaling factor $\frac{\alpha}{r}$
self.scaling = alpha / r
# Matrix $A \in \mathbb{R}^{r \times k}$
self.lora_a = nn.Parameter(torch.empty((num_embeddings, r)))
# Matrix $B \in \mathbb{R}^{d \times r}$
self.lora_b = nn.Parameter(torch.empty((r, embedding_dim)))
with torch.no_grad():
# Initialize $A$ with a normal distribution
nn.init.normal_(self.lora_a)
# Initialize $B$ to $0$ so that $\Delta W = BA$ is $0$ at initialization
nn.init.zeros_(self.lora_b)
def forward(self, x: torch.Tensor):
# Compute the embeddings $W_0 \text{onehot}(x)$
result = nn.functional.embedding(x, self.weight)
# Add $\frac{\alpha}{r} \Delta W \text{onehot}(x) = \frac{\alpha}{r} BA \text{onehot}(x_$
result += (nn.functional.embedding(x, self.lora_a) @ self.lora_b) * self.scaling
#
return result